Fix util paths

Author: szaman19

Diff for: applications/FLASK/Transformer/arg_utils.py

@@ -1,5 +1,5 @@
 import argparse
-import data_utils
+import dataset_utils
 
 
 # Command-line arguments
@@ -8,70 +8,85 @@ def add_transformer_architecture_arguments(args: argparse.Namespace):
     Adds the command line arguments to specify transformer architecture model
     parameters. This is only relevant for the encoder-decoder transformer model.
     """
-    args.add_argument('--num-attention-heads',
-                      action='store',
-                      default=8,
-                      type=int,
-                      help='number of parallel attention layers (default: 8)',
-                      metavar='NUM')
-    args.add_argument('--embed-dim',
-                      action='store',
-                      default=512,
-                      type=int,
-                      help='embedding space dimension (default: 512)',
-                      metavar='NUM')
-    args.add_argument('--feedforward-dim',
-                      action='store',
-                      default=0,
-                      type=int,
-                      help='feedforward network dimension. If zero, set to be '
-                      '4 times the embedding dimension (default: 0)',
-                      metavar='NUM')
-    args.add_argument('--num-layers',
-                      action='store',
-                      default=6,
-                      type=int,
-                      help='Number of encoder and decoder layers (default: 6)',
-                      metavar='NUM')
+    args.add_argument(
+        "--num-attention-heads",
+        action="store",
+        default=8,
+        type=int,
+        help="number of parallel attention layers (default: 8)",
+        metavar="NUM",
+    )
+    args.add_argument(
+        "--embed-dim",
+        action="store",
+        default=512,
+        type=int,
+        help="embedding space dimension (default: 512)",
+        metavar="NUM",
+    )
+    args.add_argument(
+        "--feedforward-dim",
+        action="store",
+        default=0,
+        type=int,
+        help="feedforward network dimension. If zero, set to be "
+        "4 times the embedding dimension (default: 0)",
+        metavar="NUM",
+    )
+    args.add_argument(
+        "--num-layers",
+        action="store",
+        default=6,
+        type=int,
+        help="Number of encoder and decoder layers (default: 6)",
+        metavar="NUM",
+    )
 
 
 def add_dataset_arguments(args: argparse.Namespace, default: str):
     """
     Adds dataset-related arguments to an existing argparse object.
     """
-    args.add_argument('--dataset',
-                      type=str,
-                      default=default,
-                      help=f'Which dataset to use (default: {default})',
-                      choices=data_utils.available_datasets())
-    args.add_argument('--dataset-fraction',
-                      action='store',
-                      default=1.0,
-                      type=float,
-                      help='Fraction of dataset to use (default: 1.0)',
-                      metavar='NUM')
+    args.add_argument(
+        "--dataset",
+        type=str,
+        default=default,
+        help=f"Which dataset to use (default: {default})",
+        choices=dataset_utils.available_datasets(),
+    )
+    args.add_argument(
+        "--dataset-fraction",
+        action="store",
+        default=1.0,
+        type=float,
+        help="Fraction of dataset to use (default: 1.0)",
+        metavar="NUM",
+    )
 
 
 def add_training_arguments(parser: argparse.ArgumentParser):
-    parser.add_argument("--skip-validation",
-                        action="store_true",
-                        default=False,
-                        help="Do not run validation (default: false)")
+    parser.add_argument(
+        "--skip-validation",
+        action="store_true",
+        default=False,
+        help="Do not run validation (default: false)",
+    )
     parser.add_argument(
         "--always-shuffle",
         action="store_true",
         default=False,
-        help=
-        "Always shuffle training dataset, even if pretokenized (default: false)"
+        help="Always shuffle training dataset, even if pretokenized (default: false)",
     )
     parser.add_argument(
         "--validation-set-fraction",
         type=float,
         default=0.01,
-        help="Fraction of the validation dataset to use (default: 0.001)")
+        help="Fraction of the validation dataset to use (default: 0.001)",
+    )
     parser.add_argument(
         "--save-prototext",
         action="store_true",
         default=False,
         help="Save prototext experiment file instead of protobin (slower but "
-        "debuggable) (default: false)")
+        "debuggable) (default: false)",
+    )

Diff for: applications/FLASK/Transformer/dataset_utils.py

@@ -24,24 +24,3 @@ def load_dataset(name: str):
     """
     sys.path.append(dataset_dir)
     return importlib.import_module(name)
-
-
-def add_dataset_arguments(args: argparse.Namespace, default: str):
-    """
-    Adds dataset-related arguments to an existing argparse object.
-    """
-    args.add_argument(
-        "--dataset",
-        type=str,
-        default=default,
-        help=f"Which dataset to use (default: {default})",
-        choices=available_datasets(),
-    )
-    args.add_argument(
-        "--dataset-fraction",
-        action="store",
-        default=1.0,
-        type=float,
-        help="Fraction of dataset to use (default: 1.0)",
-        metavar="NUM",
-    )

Diff for: applications/FLASK/Transformer/network.py

@@ -177,50 +177,3 @@ def _add_encoder_decoder_loss(
     # Compute cross-entropy
     ce = lbann.CrossEntropy(preds, labels, use_labels=True)
     return lbann.Scale(ce, constant=1 / (sequence_length - 1))
-
-
-def add_transformer_architecture_arguments(args: argparse.Namespace):
-    """
-    Adds the command line arguments to specify transformer architecture model
-    parameters. This is only relevant for the encoder-decoder transformer model.
-    """
-    args.add_argument(
-        "--num-attention-heads",
-        action="store",
-        default=8,
-        type=int,
-        help="number of parallel attention layers (default: 8)",
-        metavar="NUM",
-    )
-    args.add_argument(
-        "--embed-dim",
-        action="store",
-        default=512,
-        type=int,
-        help="embedding space dimension (default: 512)",
-        metavar="NUM",
-    )
-    args.add_argument(
-        "--feedforward-dim",
-        action="store",
-        default=0,
-        type=int,
-        help="feedforward network dimension. If zero, set to be "
-        "4 times the embedding dimension (default: 0)",
-        metavar="NUM",
-    )
-    args.add_argument(
-        "--num-layers",
-        action="store",
-        default=6,
-        type=int,
-        help="Number of encoder and decoder layers (default: 6)",
-        metavar="NUM",
-    )
-    args.add_argument(
-        "--positional-encoding",
-        type=str,
-        default="learned",
-        help="The type of positional encoding to use " "(default: learned)",
-        choices=[s.name.lower() for s in InputEncoding],
-    )

Diff for: applications/FLASK/Transformer/trainer.py

@@ -15,6 +15,7 @@
 import paths
 
 import dataset_utils
+import arg_utils
 import applications.FLASK.Transformer.network as network
 
 
@@ -267,85 +268,15 @@ def make_batch_script(
     return script
 
 
-def add_training_arguments(parser: argparse.ArgumentParser):
-    parser.add_argument(
-        "--skip-validation",
-        action="store_true",
-        default=False,
-        help="Do not run validation (default: false)",
-    )
-    parser.add_argument(
-        "--always-shuffle",
-        action="store_true",
-        default=False,
-        help="Always shuffle training dataset, even if pretokenized (default: false)",
-    )
-    parser.add_argument(
-        "--validation-set-fraction",
-        type=float,
-        default=0.01,
-        help="Fraction of the validation dataset to use (default: 0.001)",
-    )
-    parser.add_argument(
-        "--save-prototext",
-        action="store_true",
-        default=False,
-        help="Save prototext experiment file instead of protobin (slower but "
-        "debuggable) (default: false)",
-    )
-    parser.add_argument(
-        "--validate-every",
-        type=int,
-        default=100,
-        help="Run validation every N steps (default: 100)",
-    )
-    parser.add_argument(
-        "--random-seed",
-        action="store",
-        default=None,
-        type=int,
-        help=f"Set random seed explicitly",
-        metavar="NUM",
-    )
-    parser.add_argument(
-        "--pbar-newline-interval",
-        type=int,
-        default=100,
-        help="Number of iterations in progress bar before "
-        "printing a newline (default: 100)",
-    )
-    parser.add_argument(
-        "--pbar-width",
-        type=int,
-        default=30,
-        help="Progress bar width, if enabled (default: 30)",
-    )
-    parser.add_argument(
-        "--pbar-moving-avg",
-        type=int,
-        default=10,
-        help="Progress bar iteration time moving average "
-        "length in iterations. Disable moving average with 1 "
-        "(default: 10)",
-    )
-    parser.add_argument(
-        "--pbar-scientific",
-        action="store_true",
-        default=False,
-        help="Use scientific notation for objective value "
-        "printouts in progress bar (default: false)",
-    )
-
-
 def main():
     # Setup command line options
     parser = argparse.ArgumentParser()
     lbann.contrib.args.add_scheduler_arguments(parser, "lbann_transformer")
     lbann.contrib.args.add_profiling_arguments(parser)
     lbann.contrib.args.add_training_arguments(parser)
-    network.add_transformer_architecture_arguments(parser)
-    add_training_arguments(parser)
-    dataset_utils.add_dataset_arguments(parser, default="qm9")
+    arg_utils.add_transformer_architecture_arguments(parser)
+    arg_utils.add_training_arguments(parser)
+    arg_utils.add_dataset_arguments(parser, default="qm9")
 
     parser.add_argument(
         "--optimizer",